app.py

import gradio as gr
from huggingface_hub import InferenceClient
import pandas as pd
import io
import mysql.connector
import os

HF_API_KEY = os.getenv("hf_key")

# Hugging Face API Key (Replace with your actual key)
client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.3", token=HF_API_KEY)

def classify_task(user_input):
    """Classifies user task using Mistral."""
    prompt = f"""
You are an AI assistant that classifies user requests related to SQL and databases.  
Your task is to categorize the input into one of the following options:  

- **generate_sql** → If the user asks to generate **SQL syntax**, like SELECT, INSERT, CREATE TABLE, etc.  
- **create_table** → If the user explicitly wants to **create** a database/table on a local MySQL server.  
- **generate_demo_data_db** → If the user wants to insert demo data into a database.  
- **generate_demo_data_csv** → If the user wants to generate demo data in CSV format.  
- **analyze_data** → If the user asks for insights, trends, or statistical analysis of data.  

**Examples:**
1. "Give me SQL syntax to create a student table" → **generate_sql**  
2. "Create a student table in my database" → **create_table**  
3. "Insert some demo data in my database" → **generate_demo_data_db**  
4. "Generate sample student data in CSV format" → **generate_demo_data_csv**  
5. "Analyze student marks and trends" → **analyze_data**  

**User Input:** {user_input}  

**Output Format:** Return only the category name without any explanations.
"""

    response = client.text_generation(prompt, max_new_tokens=20).strip()
    return response
def generate_sql_query(user_input):
    """Generates SQL queries using Mistral."""
    prompt = f"Generate SQL syntax for: {user_input}"
    return client.text_generation(prompt, max_new_tokens=200).strip()
def generate_sql_query_for_create(user_input):
    """Generates SQL queries using Mistral."""
    prompt = f"""
    Generate **only** the SQL syntax for: {user_input}
    
    **Rules:**
    - No explanations, no bullet points, no extra text.
    - Return **only valid SQL**.
    
    **Example Input:**  
    "Create a student table with name, age, and email."
    
    **Example Output:**  
    ```sql
    CREATE TABLE student (
        student_id INT PRIMARY KEY AUTO_INCREMENT,
        name VARCHAR(100) NOT NULL,
        age INT NOT NULL,
        email VARCHAR(255) UNIQUE NOT NULL
    );
    ```
    """

    response = client.text_generation(prompt, max_new_tokens=200).strip()

    # Remove unnecessary text (if any)
    if "```sql" in response:
        response = response.split("```sql")[1].split("```")[0].strip()

    return response

import pymysql

def create_table(user_input, db_user, db_pass, db_host, db_name):
    try:
        # Validate inputs
        if not all([db_user, db_pass, db_host, db_name, user_input]):
            return "Please provide all required inputs (database credentials and table structure).", None

        # Generate SQL schema using Mistral
        
        schema_response = generate_sql_query_for_create(user_input)
        print(schema_response)
        # Validate schema using sqlparse
        parsed_schema = sqlparse.parse(schema_response)
        if not parsed_schema:
            return "Error: Could not generate a valid table schema.", None

        # Connect to MySQL Server
        connection = pymysql.connect(host=db_host, user=db_user, password=db_pass)
        cursor = connection.cursor()

        # Create Database if it doesn't exist
        cursor.execute(f"CREATE DATABASE IF NOT EXISTS {db_name}")
        connection.commit()
        connection.close()

        # Connect to the specified database
        connection = pymysql.connect(host=db_host, user=db_user, password=db_pass, database=db_name)
        cursor = connection.cursor()

        # Execute the generated CREATE TABLE statement
        cursor.execute(schema_response)
        connection.commit()

        return "Table created successfully.", None
    
    except pymysql.MySQLError as err:
        return f"Error: {err}", None
    
    finally:
        if 'connection' in locals() and connection.open:
            cursor.close()
            connection.close()

import mysql.connector
import re
import sqlparse  # Install via: pip install sqlparse

def generate_demo_data_db(user_input, db_user, db_pass, db_host, db_name, num_rows=10):
    """Generates and inserts structured demo data into a database using LLM."""
    
    if not all([db_user, db_pass, db_name]):
        return "Please provide database credentials.", None
    
    # Generate column definitions using LLM
    schema_prompt = f"""
    Extract column names and types from the following request:
    
    "{user_input}"
    
    **Output Format:**  
    - The first column should be an "ID" column (INTEGER, PRIMARY KEY).  
    - Provide appropriate SQL data types (VARCHAR(100) for text, INT for numbers).  
    - Use proper SQL syntax. No explanations.  

    Example Output:
    ```
    CREATE TABLE demo (
        ID INT PRIMARY KEY, 
        Name VARCHAR(100), 
        Age INT
    );
    ```
    """
    schema_response = client.text_generation(schema_prompt, max_new_tokens=200).strip()

    # Validate schema using sqlparse
    parsed_schema = sqlparse.parse(schema_response)
    if not parsed_schema:
        return "Error: Could not generate a valid table schema.", None

    # Extract table schema
    table_schema = schema_response.replace("CREATE TABLE demo (", "").replace(");", "").strip()
    
    # Connect to MySQL and create the table dynamically
    connection = mysql.connector.connect(host=db_host, user=db_user, password=db_pass, database=db_name)
    cursor = connection.cursor()
    cursor.execute(f"CREATE TABLE IF NOT EXISTS demo ({table_schema})")

    # Generate demo data using LLM
    data_prompt = f"""
    Generate {num_rows} rows of structured demo data for this table schema:
    
    ```
    {schema_response}
    ```
    
    **Output Format:**  
    - Return valid SQL INSERT statements.  
    - Ensure all values match their respective column types.  
    - Use double quotes ("") for text values.  
    - No explanations, just raw SQL.

    Example Output:
    ```
    INSERT INTO demo VALUES (1, "John Doe", 25);
    INSERT INTO demo VALUES (2, "Jane Smith", 30);
    ```
    """
    data_response = client.text_generation(data_prompt, max_new_tokens=1000).strip()

    # Extract SQL INSERT statements using a better regex
    insert_statements = re.findall(r'INSERT INTO demo VALUES \([^)]+\);', data_response, re.DOTALL)
    if not insert_statements:
        return "Error: Could not generate valid data.", None

    # Insert data into the database
    for statement in insert_statements:
        cursor.execute(statement)
    
    connection.commit()
    connection.close()
    
    return "Demo data inserted into the database successfully.", None
def generate_demo_data_csv(user_input, num_rows=10):
    """Generates realistic demo data using the LLM in valid CSV format."""
    
    prompt = f"""
    Generate a structured dataset with {num_rows} rows based on the following request:
    
    "{user_input}"
    
    **Output Format:**  
    - Ensure the response is in **valid CSV format** (comma-separated).  
    - The **first row** must be column headers.  
    - Use **double quotes for text values** to avoid formatting issues.  
    - Do **not** include explanations—just the raw CSV data.

    Example Output:
    
"ID","Name","Age","Email"
    "1","John Doe","25","[email protected]"
    "2","Jane Smith","30","[email protected]"

    """

    # Get LLM response
    response = client.text_generation(prompt, max_new_tokens=10000).strip()

    # Ensure we extract only the CSV part (some models may add explanations)
    csv_start = response.find('"ID"')  # Find where the CSV starts
    if csv_start != -1:
        response = response[csv_start:]  # Remove anything before the CSV

    # Convert to DataFrame
    try:
        df = pd.read_csv(io.StringIO(response))  # Read as CSV
    except Exception as e:
        return f"Error: Invalid CSV format. {str(e)}", None
    
    # Save to a CSV file
    file_path = "generated_data.csv"
    df.to_csv(file_path, index=False)
    
    return "Demo data generated as CSV.", file_path  # Return file path

def analyze_data(user_input):
    """Analyzes data using Mistral."""
    prompt = f"Analyze this data: {user_input}"
    return client.text_generation(prompt, max_new_tokens=200).strip()

def sql_chatbot(user_input, num_rows=10):
    task = classify_task(user_input)
    
    if "generate_sql" in task:
        return generate_sql_query(user_input), None
    
    elif "create_table" in task:
        return create_table(user_input, db_user, db_pass, db_host, db_name)
    
    elif "generate_demo_data_db" in task:
        return generate_demo_data_db(user_input, db_user, db_pass, db_host, db_name, num_rows)
    
    elif "generate_demo_data_csv" in task:
        response, file_path = generate_demo_data_csv(user_input, num_rows)
        return response, file_path
    elif "analyze_data" in task:
        return analyze_data(user_input), None
    
    return f"task:{task} \n I could not understand your request.", None

iface = gr.Interface(
    fn=sql_chatbot,
    inputs=[
        gr.Textbox(label="User Input"),
        gr.Number(label="Number of Rows", interactive=True, value=10, precision=0)
    ],
    outputs=[gr.Textbox(label="Response"), gr.File(label="File Output")]
)

iface.launch()
# print("hi")
# print(create_table("create a SQL student table","root", "123456", "localhost", "demo"))